Unsupervised Anomaly Detection of Diseases in the Female Pelvis for Real-Time MR Imaging

TL;DR

A benchmark framework for disease- and parameter-agnostic, real-time-compatible unsupervised anomaly detection in pelvic MRI is established and analysis to endometrial cancer, endometriosis, and adenomyosis is extended, revealing the influence of anatomical heterogeneity and inter-observer variability on performance interpretation.

Abstract

Pelvic diseases in women of reproductive age represent a major global health burden, with diagnosis frequently delayed due to high anatomical variability, complicating MRI interpretation. Existing AI approaches are largely disease-specific and lack real-time compatibility, limiting generalizability and clinical integration. To address these challenges, we establish a benchmark framework for disease- and parameter-agnostic, real-time-compatible unsupervised anomaly detection in pelvic MRI. The method uses a residual variational autoencoder trained exclusively on healthy sagittal T2-weighted scans acquired across diverse imaging protocols to model normal pelvic anatomy. During inference, reconstruction error heatmaps indicate deviations from learned healthy structure, enabling detection of pathological regions without labeled abnormal data. The model is trained on 294 healthy scans and augmented with diffusion-generated synthetic data to improve robustness. Quantitative evaluation on the publicly available Uterine Myoma MRI Dataset yields an average area-under-the-curve (AUC) value of 0.736, with 0.828 sensitivity and 0.692 specificity. Additional inter-observer clinical evaluation extends analysis to endometrial cancer, endometriosis, and adenomyosis, revealing the influence of anatomical heterogeneity and inter-observer variability on performance interpretation. With a reconstruction time of approximately 92.6 frames per second, the proposed framework establishes a baseline for unsupervised anomaly detection in the female pelvis and supports future integration into real-time MRI. Code is available upon request (https://github.com/AniKnu/UADPelvis), prospective data sets are available for academic collaboration.

Figures (8)

• Figure 1: The upper rows depict healthy uteri, while the lower rows represent pathological cases, with arrows indicating anomalies such as myomas, cysts, and endometriosis, highlighting the anatomical variability of the female pelvis.
• Figure 2: Overview of the proposed unsupervised pelvic MRI anomaly detection pipeline. Preprocessed sagittal T2w scans are augmented using an in-house DDPM and used to train a residual variational autoencoder on healthy anatomy. During inference, reconstruction errors are post-processed and overlaid to highlight anomalies. Evaluation is based on ground-truth segmentations.
• Figure 3: Pre-processing pipeline for standardized model input, including voxel and intensity normalization, anatomical segmentation using an attention-based 3D U-Net tripathydeep, and segmentation-driven bounding box cropping.
• Figure 4: Residual variational autoencoder architecture with four residual blocks in each encoder and decoder, with a latent space dimension of $256$.
• Figure 5: Random synthetic samples illustrating uterine orientations (blue: RV, AF, green: RV, RF), image artifacts (orange), and Nabothian cysts (red arrows).